1/*
   2 *  gendisk handling
   3 */
   4
   5#include <linux/module.h>
   6#include <linux/fs.h>
   7#include <linux/genhd.h>
   8#include <linux/kdev_t.h>
   9#include <linux/kernel.h>
  10#include <linux/blkdev.h>
  11#include <linux/backing-dev.h>
  12#include <linux/init.h>
  13#include <linux/spinlock.h>
  14#include <linux/proc_fs.h>
  15#include <linux/seq_file.h>
  16#include <linux/slab.h>
  17#include <linux/kmod.h>
  18#include <linux/kobj_map.h>
  19#include <linux/mutex.h>
  20#include <linux/idr.h>
  21#include <linux/log2.h>
  22#include <linux/pm_runtime.h>
  23#include <linux/badblocks.h>
  24
  25#include "blk.h"
  26
  27static DEFINE_MUTEX(block_class_lock);
  28struct kobject *block_depr;
  29
  30/* for extended dynamic devt allocation, currently only one major is used */
  31#define NR_EXT_DEVT		(1 << MINORBITS)
  32
  33/* For extended devt allocation.  ext_devt_lock prevents look up
  34 * results from going away underneath its user.
  35 */
  36static DEFINE_SPINLOCK(ext_devt_lock);
  37static DEFINE_IDR(ext_devt_idr);
  38
  39static const struct device_type disk_type;
  40
  41static void disk_check_events(struct disk_events *ev,
  42			      unsigned int *clearing_ptr);
  43static void disk_alloc_events(struct gendisk *disk);
  44static void disk_add_events(struct gendisk *disk);
  45static void disk_del_events(struct gendisk *disk);
  46static void disk_release_events(struct gendisk *disk);
  47
  48void part_inc_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
  49{
  50	if (q->mq_ops)
  51		return;
  52
  53	atomic_inc(&part->in_flight[rw]);
  54	if (part->partno)
  55		atomic_inc(&part_to_disk(part)->part0.in_flight[rw]);
  56}
  57
  58void part_dec_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
  59{
  60	if (q->mq_ops)
  61		return;
  62
  63	atomic_dec(&part->in_flight[rw]);
  64	if (part->partno)
  65		atomic_dec(&part_to_disk(part)->part0.in_flight[rw]);
  66}
  67
  68void part_in_flight(struct request_queue *q, struct hd_struct *part,
  69		    unsigned int inflight[2])
  70{
  71	if (q->mq_ops) {
  72		blk_mq_in_flight(q, part, inflight);
  73		return;
  74	}
  75
  76	inflight[0] = atomic_read(&part->in_flight[0]) +
  77			atomic_read(&part->in_flight[1]);
  78	if (part->partno) {
  79		part = &part_to_disk(part)->part0;
  80		inflight[1] = atomic_read(&part->in_flight[0]) +
  81				atomic_read(&part->in_flight[1]);
  82	}
  83}
  84
  85void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
  86		       unsigned int inflight[2])
  87{
  88	if (q->mq_ops) {
  89		blk_mq_in_flight_rw(q, part, inflight);
  90		return;
  91	}
  92
  93	inflight[0] = atomic_read(&part->in_flight[0]);
  94	inflight[1] = atomic_read(&part->in_flight[1]);
  95}
  96
  97struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
  98{
  99	struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
 100
 101	if (unlikely(partno < 0 || partno >= ptbl->len))
 102		return NULL;
 103	return rcu_dereference(ptbl->part[partno]);
 104}
 105
 106/**
 107 * disk_get_part - get partition
 108 * @disk: disk to look partition from
 109 * @partno: partition number
 110 *
 111 * Look for partition @partno from @disk.  If found, increment
 112 * reference count and return it.
 113 *
 114 * CONTEXT:
 115 * Don't care.
 116 *
 117 * RETURNS:
 118 * Pointer to the found partition on success, NULL if not found.
 119 */
 120struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
 121{
 122	struct hd_struct *part;
 
 
 
 
 123
 124	rcu_read_lock();
 125	part = __disk_get_part(disk, partno);
 126	if (part)
 127		get_device(part_to_dev(part));
 
 
 
 
 
 128	rcu_read_unlock();
 129
 130	return part;
 131}
 132EXPORT_SYMBOL_GPL(disk_get_part);
 133
 134/**
 135 * disk_part_iter_init - initialize partition iterator
 136 * @piter: iterator to initialize
 137 * @disk: disk to iterate over
 138 * @flags: DISK_PITER_* flags
 139 *
 140 * Initialize @piter so that it iterates over partitions of @disk.
 141 *
 142 * CONTEXT:
 143 * Don't care.
 144 */
 145void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
 146			  unsigned int flags)
 147{
 148	struct disk_part_tbl *ptbl;
 149
 150	rcu_read_lock();
 151	ptbl = rcu_dereference(disk->part_tbl);
 152
 153	piter->disk = disk;
 154	piter->part = NULL;
 155
 156	if (flags & DISK_PITER_REVERSE)
 157		piter->idx = ptbl->len - 1;
 158	else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
 159		piter->idx = 0;
 160	else
 161		piter->idx = 1;
 162
 163	piter->flags = flags;
 164
 165	rcu_read_unlock();
 166}
 167EXPORT_SYMBOL_GPL(disk_part_iter_init);
 168
 169/**
 170 * disk_part_iter_next - proceed iterator to the next partition and return it
 171 * @piter: iterator of interest
 172 *
 173 * Proceed @piter to the next partition and return it.
 174 *
 175 * CONTEXT:
 176 * Don't care.
 177 */
 178struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 179{
 180	struct disk_part_tbl *ptbl;
 181	int inc, end;
 182
 183	/* put the last partition */
 184	disk_put_part(piter->part);
 185	piter->part = NULL;
 186
 187	/* get part_tbl */
 188	rcu_read_lock();
 189	ptbl = rcu_dereference(piter->disk->part_tbl);
 190
 191	/* determine iteration parameters */
 192	if (piter->flags & DISK_PITER_REVERSE) {
 193		inc = -1;
 194		if (piter->flags & (DISK_PITER_INCL_PART0 |
 195				    DISK_PITER_INCL_EMPTY_PART0))
 196			end = -1;
 197		else
 198			end = 0;
 199	} else {
 200		inc = 1;
 201		end = ptbl->len;
 202	}
 203
 204	/* iterate to the next partition */
 205	for (; piter->idx != end; piter->idx += inc) {
 206		struct hd_struct *part;
 207
 208		part = rcu_dereference(ptbl->part[piter->idx]);
 209		if (!part)
 210			continue;
 211		if (!part_nr_sects_read(part) &&
 212		    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
 213		    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
 214		      piter->idx == 0))
 215			continue;
 216
 217		get_device(part_to_dev(part));
 218		piter->part = part;
 219		piter->idx += inc;
 220		break;
 221	}
 222
 223	rcu_read_unlock();
 224
 225	return piter->part;
 226}
 227EXPORT_SYMBOL_GPL(disk_part_iter_next);
 228
 229/**
 230 * disk_part_iter_exit - finish up partition iteration
 231 * @piter: iter of interest
 232 *
 233 * Called when iteration is over.  Cleans up @piter.
 234 *
 235 * CONTEXT:
 236 * Don't care.
 237 */
 238void disk_part_iter_exit(struct disk_part_iter *piter)
 239{
 240	disk_put_part(piter->part);
 241	piter->part = NULL;
 242}
 243EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 244
 245static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 246{
 247	return part->start_sect <= sector &&
 248		sector < part->start_sect + part_nr_sects_read(part);
 249}
 250
 251/**
 252 * disk_map_sector_rcu - map sector to partition
 253 * @disk: gendisk of interest
 254 * @sector: sector to map
 255 *
 256 * Find out which partition @sector maps to on @disk.  This is
 257 * primarily used for stats accounting.
 258 *
 259 * CONTEXT:
 260 * RCU read locked.  The returned partition pointer is valid only
 261 * while preemption is disabled.
 262 *
 263 * RETURNS:
 264 * Found partition on success, part0 is returned if no partition matches
 265 */
 266struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 267{
 268	struct disk_part_tbl *ptbl;
 269	struct hd_struct *part;
 270	int i;
 271
 272	ptbl = rcu_dereference(disk->part_tbl);
 273
 274	part = rcu_dereference(ptbl->last_lookup);
 275	if (part && sector_in_part(part, sector))
 276		return part;
 277
 278	for (i = 1; i < ptbl->len; i++) {
 279		part = rcu_dereference(ptbl->part[i]);
 280
 281		if (part && sector_in_part(part, sector)) {
 282			rcu_assign_pointer(ptbl->last_lookup, part);
 283			return part;
 284		}
 285	}
 286	return &disk->part0;
 287}
 288EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
 289
 290/*
 291 * Can be deleted altogether. Later.
 292 *
 293 */
 294#define BLKDEV_MAJOR_HASH_SIZE 255
 295static struct blk_major_name {
 296	struct blk_major_name *next;
 297	int major;
 298	char name[16];
 299} *major_names[BLKDEV_MAJOR_HASH_SIZE];
 300
 301/* index in the above - for now: assume no multimajor ranges */
 302static inline int major_to_index(unsigned major)
 303{
 304	return major % BLKDEV_MAJOR_HASH_SIZE;
 305}
 306
 307#ifdef CONFIG_PROC_FS
 308void blkdev_show(struct seq_file *seqf, off_t offset)
 309{
 310	struct blk_major_name *dp;
 311
 312	mutex_lock(&block_class_lock);
 313	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
 314		if (dp->major == offset)
 315			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 316	mutex_unlock(&block_class_lock);
 
 317}
 318#endif /* CONFIG_PROC_FS */
 319
 320/**
 321 * register_blkdev - register a new block device
 322 *
 323 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
 324 *         @major = 0, try to allocate any unused major number.
 325 * @name: the name of the new block device as a zero terminated string
 326 *
 327 * The @name must be unique within the system.
 328 *
 329 * The return value depends on the @major input parameter:
 330 *
 331 *  - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
 332 *    then the function returns zero on success, or a negative error code
 333 *  - if any unused major number was requested with @major = 0 parameter
 334 *    then the return value is the allocated major number in range
 335 *    [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
 336 *
 337 * See Documentation/admin-guide/devices.txt for the list of allocated
 338 * major numbers.
 339 */
 340int register_blkdev(unsigned int major, const char *name)
 341{
 342	struct blk_major_name **n, *p;
 343	int index, ret = 0;
 344
 345	mutex_lock(&block_class_lock);
 346
 347	/* temporary */
 348	if (major == 0) {
 349		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 350			if (major_names[index] == NULL)
 351				break;
 352		}
 353
 354		if (index == 0) {
 355			printk("register_blkdev: failed to get major for %s\n",
 356			       name);
 357			ret = -EBUSY;
 358			goto out;
 359		}
 360		major = index;
 361		ret = major;
 362	}
 363
 364	if (major >= BLKDEV_MAJOR_MAX) {
 365		pr_err("register_blkdev: major requested (%u) is greater than the maximum (%u) for %s\n",
 366		       major, BLKDEV_MAJOR_MAX-1, name);
 367
 368		ret = -EINVAL;
 369		goto out;
 370	}
 371
 372	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 373	if (p == NULL) {
 374		ret = -ENOMEM;
 375		goto out;
 376	}
 377
 378	p->major = major;
 379	strlcpy(p->name, name, sizeof(p->name));
 380	p->next = NULL;
 381	index = major_to_index(major);
 382
 383	for (n = &major_names[index]; *n; n = &(*n)->next) {
 384		if ((*n)->major == major)
 385			break;
 386	}
 387	if (!*n)
 388		*n = p;
 389	else
 390		ret = -EBUSY;
 391
 392	if (ret < 0) {
 393		printk("register_blkdev: cannot get major %u for %s\n",
 394		       major, name);
 395		kfree(p);
 396	}
 397out:
 398	mutex_unlock(&block_class_lock);
 399	return ret;
 400}
 401
 402EXPORT_SYMBOL(register_blkdev);
 403
 404void unregister_blkdev(unsigned int major, const char *name)
 405{
 406	struct blk_major_name **n;
 407	struct blk_major_name *p = NULL;
 408	int index = major_to_index(major);
 409
 410	mutex_lock(&block_class_lock);
 411	for (n = &major_names[index]; *n; n = &(*n)->next)
 412		if ((*n)->major == major)
 413			break;
 414	if (!*n || strcmp((*n)->name, name)) {
 415		WARN_ON(1);
 416	} else {
 417		p = *n;
 418		*n = p->next;
 419	}
 420	mutex_unlock(&block_class_lock);
 421	kfree(p);
 422}
 423
 424EXPORT_SYMBOL(unregister_blkdev);
 425
 426static struct kobj_map *bdev_map;
 427
 428/**
 429 * blk_mangle_minor - scatter minor numbers apart
 430 * @minor: minor number to mangle
 431 *
 432 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
 433 * is enabled.  Mangling twice gives the original value.
 434 *
 435 * RETURNS:
 436 * Mangled value.
 437 *
 438 * CONTEXT:
 439 * Don't care.
 440 */
 441static int blk_mangle_minor(int minor)
 442{
 443#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 444	int i;
 445
 446	for (i = 0; i < MINORBITS / 2; i++) {
 447		int low = minor & (1 << i);
 448		int high = minor & (1 << (MINORBITS - 1 - i));
 449		int distance = MINORBITS - 1 - 2 * i;
 450
 451		minor ^= low | high;	/* clear both bits */
 452		low <<= distance;	/* swap the positions */
 453		high >>= distance;
 454		minor |= low | high;	/* and set */
 455	}
 456#endif
 457	return minor;
 458}
 459
 460/**
 461 * blk_alloc_devt - allocate a dev_t for a partition
 462 * @part: partition to allocate dev_t for
 463 * @devt: out parameter for resulting dev_t
 464 *
 465 * Allocate a dev_t for block device.
 466 *
 467 * RETURNS:
 468 * 0 on success, allocated dev_t is returned in *@devt.  -errno on
 469 * failure.
 470 *
 471 * CONTEXT:
 472 * Might sleep.
 473 */
 474int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 475{
 476	struct gendisk *disk = part_to_disk(part);
 477	int idx;
 478
 479	/* in consecutive minor range? */
 480	if (part->partno < disk->minors) {
 481		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
 482		return 0;
 483	}
 484
 485	/* allocate ext devt */
 486	idr_preload(GFP_KERNEL);
 487
 488	spin_lock_bh(&ext_devt_lock);
 489	idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
 490	spin_unlock_bh(&ext_devt_lock);
 491
 492	idr_preload_end();
 493	if (idx < 0)
 494		return idx == -ENOSPC ? -EBUSY : idx;
 495
 496	*devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
 497	return 0;
 498}
 499
 500/**
 501 * blk_free_devt - free a dev_t
 502 * @devt: dev_t to free
 503 *
 504 * Free @devt which was allocated using blk_alloc_devt().
 505 *
 506 * CONTEXT:
 507 * Might sleep.
 508 */
 509void blk_free_devt(dev_t devt)
 510{
 511	if (devt == MKDEV(0, 0))
 512		return;
 513
 514	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
 515		spin_lock_bh(&ext_devt_lock);
 516		idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 517		spin_unlock_bh(&ext_devt_lock);
 518	}
 519}
 520
 521static char *bdevt_str(dev_t devt, char *buf)
 522{
 523	if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
 524		char tbuf[BDEVT_SIZE];
 525		snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
 526		snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
 527	} else
 528		snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
 529
 530	return buf;
 531}
 532
 533/*
 534 * Register device numbers dev..(dev+range-1)
 535 * range must be nonzero
 536 * The hash chain is sorted on range, so that subranges can override.
 537 */
 538void blk_register_region(dev_t devt, unsigned long range, struct module *module,
 539			 struct kobject *(*probe)(dev_t, int *, void *),
 540			 int (*lock)(dev_t, void *), void *data)
 541{
 542	kobj_map(bdev_map, devt, range, module, probe, lock, data);
 543}
 544
 545EXPORT_SYMBOL(blk_register_region);
 546
 547void blk_unregister_region(dev_t devt, unsigned long range)
 548{
 549	kobj_unmap(bdev_map, devt, range);
 550}
 551
 552EXPORT_SYMBOL(blk_unregister_region);
 553
 554static struct kobject *exact_match(dev_t devt, int *partno, void *data)
 555{
 556	struct gendisk *p = data;
 557
 558	return &disk_to_dev(p)->kobj;
 559}
 560
 561static int exact_lock(dev_t devt, void *data)
 562{
 563	struct gendisk *p = data;
 564
 565	if (!get_disk_and_module(p))
 566		return -1;
 567	return 0;
 568}
 569
 570static void register_disk(struct device *parent, struct gendisk *disk)
 571{
 572	struct device *ddev = disk_to_dev(disk);
 573	struct block_device *bdev;
 574	struct disk_part_iter piter;
 575	struct hd_struct *part;
 576	int err;
 577
 578	ddev->parent = parent;
 579
 580	dev_set_name(ddev, "%s", disk->disk_name);
 581
 582	/* delay uevents, until we scanned partition table */
 583	dev_set_uevent_suppress(ddev, 1);
 584
 585	if (device_add(ddev))
 586		return;
 587	if (!sysfs_deprecated) {
 588		err = sysfs_create_link(block_depr, &ddev->kobj,
 589					kobject_name(&ddev->kobj));
 590		if (err) {
 591			device_del(ddev);
 592			return;
 593		}
 594	}
 595
 596	/*
 597	 * avoid probable deadlock caused by allocating memory with
 598	 * GFP_KERNEL in runtime_resume callback of its all ancestor
 599	 * devices
 600	 */
 601	pm_runtime_set_memalloc_noio(ddev, true);
 602
 603	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
 604	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 605
 606	if (disk->flags & GENHD_FL_HIDDEN) {
 607		dev_set_uevent_suppress(ddev, 0);
 608		return;
 609	}
 610
 611	/* No minors to use for partitions */
 612	if (!disk_part_scan_enabled(disk))
 613		goto exit;
 614
 615	/* No such device (e.g., media were just removed) */
 616	if (!get_capacity(disk))
 617		goto exit;
 618
 619	bdev = bdget_disk(disk, 0);
 620	if (!bdev)
 621		goto exit;
 622
 623	bdev->bd_invalidated = 1;
 624	err = blkdev_get(bdev, FMODE_READ, NULL);
 625	if (err < 0)
 626		goto exit;
 627	blkdev_put(bdev, FMODE_READ);
 628
 629exit:
 630	/* announce disk after possible partitions are created */
 631	dev_set_uevent_suppress(ddev, 0);
 632	kobject_uevent(&ddev->kobj, KOBJ_ADD);
 633
 634	/* announce possible partitions */
 635	disk_part_iter_init(&piter, disk, 0);
 636	while ((part = disk_part_iter_next(&piter)))
 637		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
 638	disk_part_iter_exit(&piter);
 639
 640	err = sysfs_create_link(&ddev->kobj,
 641				&disk->queue->backing_dev_info->dev->kobj,
 642				"bdi");
 643	WARN_ON(err);
 644}
 645
 646/**
 647 * __device_add_disk - add disk information to kernel list
 648 * @parent: parent device for the disk
 649 * @disk: per-device partitioning information
 650 * @register_queue: register the queue if set to true
 651 *
 652 * This function registers the partitioning information in @disk
 653 * with the kernel.
 654 *
 655 * FIXME: error handling
 656 */
 657static void __device_add_disk(struct device *parent, struct gendisk *disk,
 658			      bool register_queue)
 659{
 
 660	dev_t devt;
 661	int retval;
 662
 663	/* minors == 0 indicates to use ext devt from part0 and should
 664	 * be accompanied with EXT_DEVT flag.  Make sure all
 665	 * parameters make sense.
 666	 */
 667	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
 668	WARN_ON(!disk->minors &&
 669		!(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
 670
 671	disk->flags |= GENHD_FL_UP;
 672
 673	retval = blk_alloc_devt(&disk->part0, &devt);
 674	if (retval) {
 675		WARN_ON(1);
 676		return;
 677	}
 
 
 
 
 
 678	disk->major = MAJOR(devt);
 679	disk->first_minor = MINOR(devt);
 680
 681	disk_alloc_events(disk);
 682
 683	if (disk->flags & GENHD_FL_HIDDEN) {
 684		/*
 685		 * Don't let hidden disks show up in /proc/partitions,
 686		 * and don't bother scanning for partitions either.
 687		 */
 688		disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 689		disk->flags |= GENHD_FL_NO_PART_SCAN;
 690	} else {
 691		int ret;
 692
 693		/* Register BDI before referencing it from bdev */
 694		disk_to_dev(disk)->devt = devt;
 695		ret = bdi_register_owner(disk->queue->backing_dev_info,
 696						disk_to_dev(disk));
 697		WARN_ON(ret);
 698		blk_register_region(disk_devt(disk), disk->minors, NULL,
 699				    exact_match, exact_lock, disk);
 700	}
 701	register_disk(parent, disk);
 702	if (register_queue)
 703		blk_register_queue(disk);
 704
 705	/*
 706	 * Take an extra ref on queue which will be put on disk_release()
 707	 * so that it sticks around as long as @disk is there.
 708	 */
 709	WARN_ON_ONCE(!blk_get_queue(disk->queue));
 710
 
 
 
 
 711	disk_add_events(disk);
 712	blk_integrity_add(disk);
 713}
 714
 715void device_add_disk(struct device *parent, struct gendisk *disk)
 716{
 717	__device_add_disk(parent, disk, true);
 718}
 719EXPORT_SYMBOL(device_add_disk);
 720
 721void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
 722{
 723	__device_add_disk(parent, disk, false);
 724}
 725EXPORT_SYMBOL(device_add_disk_no_queue_reg);
 726
 727void del_gendisk(struct gendisk *disk)
 728{
 729	struct disk_part_iter piter;
 730	struct hd_struct *part;
 731
 732	blk_integrity_del(disk);
 733	disk_del_events(disk);
 734
 735	/*
 736	 * Block lookups of the disk until all bdevs are unhashed and the
 737	 * disk is marked as dead (GENHD_FL_UP cleared).
 738	 */
 739	down_write(&disk->lookup_sem);
 740	/* invalidate stuff */
 741	disk_part_iter_init(&piter, disk,
 742			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 743	while ((part = disk_part_iter_next(&piter))) {
 744		invalidate_partition(disk, part->partno);
 745		bdev_unhash_inode(part_devt(part));
 746		delete_partition(disk, part->partno);
 747	}
 748	disk_part_iter_exit(&piter);
 749
 750	invalidate_partition(disk, 0);
 751	bdev_unhash_inode(disk_devt(disk));
 752	set_capacity(disk, 0);
 753	disk->flags &= ~GENHD_FL_UP;
 754	up_write(&disk->lookup_sem);
 755
 756	if (!(disk->flags & GENHD_FL_HIDDEN))
 757		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 758	if (disk->queue) {
 759		/*
 760		 * Unregister bdi before releasing device numbers (as they can
 761		 * get reused and we'd get clashes in sysfs).
 762		 */
 763		if (!(disk->flags & GENHD_FL_HIDDEN))
 764			bdi_unregister(disk->queue->backing_dev_info);
 765		blk_unregister_queue(disk);
 766	} else {
 767		WARN_ON(1);
 768	}
 769
 770	if (!(disk->flags & GENHD_FL_HIDDEN))
 771		blk_unregister_region(disk_devt(disk), disk->minors);
 772
 773	kobject_put(disk->part0.holder_dir);
 774	kobject_put(disk->slave_dir);
 775
 776	part_stat_set_all(&disk->part0, 0);
 777	disk->part0.stamp = 0;
 778	if (!sysfs_deprecated)
 779		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 780	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
 781	device_del(disk_to_dev(disk));
 782}
 783EXPORT_SYMBOL(del_gendisk);
 784
 785/* sysfs access to bad-blocks list. */
 786static ssize_t disk_badblocks_show(struct device *dev,
 787					struct device_attribute *attr,
 788					char *page)
 789{
 790	struct gendisk *disk = dev_to_disk(dev);
 791
 792	if (!disk->bb)
 793		return sprintf(page, "\n");
 794
 795	return badblocks_show(disk->bb, page, 0);
 796}
 797
 798static ssize_t disk_badblocks_store(struct device *dev,
 799					struct device_attribute *attr,
 800					const char *page, size_t len)
 801{
 802	struct gendisk *disk = dev_to_disk(dev);
 803
 804	if (!disk->bb)
 805		return -ENXIO;
 806
 807	return badblocks_store(disk->bb, page, len, 0);
 808}
 809
 810/**
 811 * get_gendisk - get partitioning information for a given device
 812 * @devt: device to get partitioning information for
 813 * @partno: returned partition index
 814 *
 815 * This function gets the structure containing partitioning
 816 * information for the given device @devt.
 817 */
 818struct gendisk *get_gendisk(dev_t devt, int *partno)
 819{
 820	struct gendisk *disk = NULL;
 821
 822	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 823		struct kobject *kobj;
 824
 825		kobj = kobj_lookup(bdev_map, devt, partno);
 826		if (kobj)
 827			disk = dev_to_disk(kobj_to_dev(kobj));
 828	} else {
 829		struct hd_struct *part;
 830
 831		spin_lock_bh(&ext_devt_lock);
 832		part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 833		if (part && get_disk_and_module(part_to_disk(part))) {
 834			*partno = part->partno;
 835			disk = part_to_disk(part);
 836		}
 837		spin_unlock_bh(&ext_devt_lock);
 838	}
 839
 840	if (!disk)
 841		return NULL;
 842
 843	/*
 844	 * Synchronize with del_gendisk() to not return disk that is being
 845	 * destroyed.
 846	 */
 847	down_read(&disk->lookup_sem);
 848	if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
 849		     !(disk->flags & GENHD_FL_UP))) {
 850		up_read(&disk->lookup_sem);
 851		put_disk_and_module(disk);
 852		disk = NULL;
 853	} else {
 854		up_read(&disk->lookup_sem);
 855	}
 856	return disk;
 857}
 858EXPORT_SYMBOL(get_gendisk);
 859
 860/**
 861 * bdget_disk - do bdget() by gendisk and partition number
 862 * @disk: gendisk of interest
 863 * @partno: partition number
 864 *
 865 * Find partition @partno from @disk, do bdget() on it.
 866 *
 867 * CONTEXT:
 868 * Don't care.
 869 *
 870 * RETURNS:
 871 * Resulting block_device on success, NULL on failure.
 872 */
 873struct block_device *bdget_disk(struct gendisk *disk, int partno)
 874{
 875	struct hd_struct *part;
 876	struct block_device *bdev = NULL;
 877
 878	part = disk_get_part(disk, partno);
 879	if (part)
 880		bdev = bdget(part_devt(part));
 881	disk_put_part(part);
 882
 883	return bdev;
 884}
 885EXPORT_SYMBOL(bdget_disk);
 886
 887/*
 888 * print a full list of all partitions - intended for places where the root
 889 * filesystem can't be mounted and thus to give the victim some idea of what
 890 * went wrong
 891 */
 892void __init printk_all_partitions(void)
 893{
 894	struct class_dev_iter iter;
 895	struct device *dev;
 896
 897	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 898	while ((dev = class_dev_iter_next(&iter))) {
 899		struct gendisk *disk = dev_to_disk(dev);
 900		struct disk_part_iter piter;
 901		struct hd_struct *part;
 902		char name_buf[BDEVNAME_SIZE];
 903		char devt_buf[BDEVT_SIZE];
 904
 905		/*
 906		 * Don't show empty devices or things that have been
 907		 * suppressed
 908		 */
 909		if (get_capacity(disk) == 0 ||
 910		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 911			continue;
 912
 913		/*
 914		 * Note, unlike /proc/partitions, I am showing the
 915		 * numbers in hex - the same format as the root=
 916		 * option takes.
 917		 */
 918		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 919		while ((part = disk_part_iter_next(&piter))) {
 920			bool is_part0 = part == &disk->part0;
 921
 922			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 923			       bdevt_str(part_devt(part), devt_buf),
 924			       (unsigned long long)part_nr_sects_read(part) >> 1
 925			       , disk_name(disk, part->partno, name_buf),
 926			       part->info ? part->info->uuid : "");
 927			if (is_part0) {
 928				if (dev->parent && dev->parent->driver)
 929					printk(" driver: %s\n",
 930					      dev->parent->driver->name);
 931				else
 932					printk(" (driver?)\n");
 933			} else
 934				printk("\n");
 935		}
 936		disk_part_iter_exit(&piter);
 937	}
 938	class_dev_iter_exit(&iter);
 939}
 940
 941#ifdef CONFIG_PROC_FS
 942/* iterator */
 943static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 944{
 945	loff_t skip = *pos;
 946	struct class_dev_iter *iter;
 947	struct device *dev;
 948
 949	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 950	if (!iter)
 951		return ERR_PTR(-ENOMEM);
 952
 953	seqf->private = iter;
 954	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 955	do {
 956		dev = class_dev_iter_next(iter);
 957		if (!dev)
 958			return NULL;
 959	} while (skip--);
 960
 961	return dev_to_disk(dev);
 962}
 963
 964static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 965{
 966	struct device *dev;
 967
 968	(*pos)++;
 969	dev = class_dev_iter_next(seqf->private);
 970	if (dev)
 971		return dev_to_disk(dev);
 972
 973	return NULL;
 974}
 975
 976static void disk_seqf_stop(struct seq_file *seqf, void *v)
 977{
 978	struct class_dev_iter *iter = seqf->private;
 979
 980	/* stop is called even after start failed :-( */
 981	if (iter) {
 982		class_dev_iter_exit(iter);
 983		kfree(iter);
 984		seqf->private = NULL;
 985	}
 986}
 987
 988static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 989{
 990	void *p;
 991
 992	p = disk_seqf_start(seqf, pos);
 993	if (!IS_ERR_OR_NULL(p) && !*pos)
 994		seq_puts(seqf, "major minor  #blocks  name\n\n");
 995	return p;
 996}
 997
 998static int show_partition(struct seq_file *seqf, void *v)
 999{
1000	struct gendisk *sgp = v;
1001	struct disk_part_iter piter;
1002	struct hd_struct *part;
1003	char buf[BDEVNAME_SIZE];
1004
1005	/* Don't show non-partitionable removeable devices or empty devices */
1006	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1007				   (sgp->flags & GENHD_FL_REMOVABLE)))
1008		return 0;
1009	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1010		return 0;
1011
1012	/* show the full disk and all non-0 size partitions of it */
1013	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1014	while ((part = disk_part_iter_next(&piter)))
1015		seq_printf(seqf, "%4d  %7d %10llu %s\n",
1016			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
1017			   (unsigned long long)part_nr_sects_read(part) >> 1,
1018			   disk_name(sgp, part->partno, buf));
1019	disk_part_iter_exit(&piter);
1020
1021	return 0;
1022}
1023
1024static const struct seq_operations partitions_op = {
1025	.start	= show_partition_start,
1026	.next	= disk_seqf_next,
1027	.stop	= disk_seqf_stop,
1028	.show	= show_partition
1029};
1030
1031static int partitions_open(struct inode *inode, struct file *file)
1032{
1033	return seq_open(file, &partitions_op);
1034}
1035
1036static const struct file_operations proc_partitions_operations = {
1037	.open		= partitions_open,
1038	.read		= seq_read,
1039	.llseek		= seq_lseek,
1040	.release	= seq_release,
1041};
1042#endif
1043
1044
1045static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1046{
1047	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1048		/* Make old-style 2.4 aliases work */
1049		request_module("block-major-%d", MAJOR(devt));
1050	return NULL;
1051}
1052
1053static int __init genhd_device_init(void)
1054{
1055	int error;
1056
1057	block_class.dev_kobj = sysfs_dev_block_kobj;
1058	error = class_register(&block_class);
1059	if (unlikely(error))
1060		return error;
1061	bdev_map = kobj_map_init(base_probe, &block_class_lock);
1062	blk_dev_init();
1063
1064	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1065
1066	/* create top-level block dir */
1067	if (!sysfs_deprecated)
1068		block_depr = kobject_create_and_add("block", NULL);
1069	return 0;
1070}
1071
1072subsys_initcall(genhd_device_init);
1073
1074static ssize_t disk_range_show(struct device *dev,
1075			       struct device_attribute *attr, char *buf)
1076{
1077	struct gendisk *disk = dev_to_disk(dev);
1078
1079	return sprintf(buf, "%d\n", disk->minors);
1080}
1081
1082static ssize_t disk_ext_range_show(struct device *dev,
1083				   struct device_attribute *attr, char *buf)
1084{
1085	struct gendisk *disk = dev_to_disk(dev);
1086
1087	return sprintf(buf, "%d\n", disk_max_parts(disk));
1088}
1089
1090static ssize_t disk_removable_show(struct device *dev,
1091				   struct device_attribute *attr, char *buf)
1092{
1093	struct gendisk *disk = dev_to_disk(dev);
1094
1095	return sprintf(buf, "%d\n",
1096		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1097}
1098
1099static ssize_t disk_hidden_show(struct device *dev,
1100				   struct device_attribute *attr, char *buf)
1101{
1102	struct gendisk *disk = dev_to_disk(dev);
1103
1104	return sprintf(buf, "%d\n",
1105		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1106}
1107
1108static ssize_t disk_ro_show(struct device *dev,
1109				   struct device_attribute *attr, char *buf)
1110{
1111	struct gendisk *disk = dev_to_disk(dev);
1112
1113	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1114}
1115
1116static ssize_t disk_capability_show(struct device *dev,
1117				    struct device_attribute *attr, char *buf)
1118{
1119	struct gendisk *disk = dev_to_disk(dev);
1120
1121	return sprintf(buf, "%x\n", disk->flags);
1122}
1123
1124static ssize_t disk_alignment_offset_show(struct device *dev,
1125					  struct device_attribute *attr,
1126					  char *buf)
1127{
1128	struct gendisk *disk = dev_to_disk(dev);
1129
1130	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1131}
1132
1133static ssize_t disk_discard_alignment_show(struct device *dev,
1134					   struct device_attribute *attr,
1135					   char *buf)
1136{
1137	struct gendisk *disk = dev_to_disk(dev);
1138
1139	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1140}
1141
1142static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1143static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1144static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1145static DEVICE_ATTR(hidden, S_IRUGO, disk_hidden_show, NULL);
1146static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1147static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1148static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1149static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1150		   NULL);
1151static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1152static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1153static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1154static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1155		disk_badblocks_store);
1156#ifdef CONFIG_FAIL_MAKE_REQUEST
1157static struct device_attribute dev_attr_fail =
1158	__ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1159#endif
1160#ifdef CONFIG_FAIL_IO_TIMEOUT
1161static struct device_attribute dev_attr_fail_timeout =
1162	__ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
1163		part_timeout_store);
1164#endif
1165
1166static struct attribute *disk_attrs[] = {
1167	&dev_attr_range.attr,
1168	&dev_attr_ext_range.attr,
1169	&dev_attr_removable.attr,
1170	&dev_attr_hidden.attr,
1171	&dev_attr_ro.attr,
1172	&dev_attr_size.attr,
1173	&dev_attr_alignment_offset.attr,
1174	&dev_attr_discard_alignment.attr,
1175	&dev_attr_capability.attr,
1176	&dev_attr_stat.attr,
1177	&dev_attr_inflight.attr,
1178	&dev_attr_badblocks.attr,
1179#ifdef CONFIG_FAIL_MAKE_REQUEST
1180	&dev_attr_fail.attr,
1181#endif
1182#ifdef CONFIG_FAIL_IO_TIMEOUT
1183	&dev_attr_fail_timeout.attr,
1184#endif
1185	NULL
1186};
1187
1188static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1189{
1190	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1191	struct gendisk *disk = dev_to_disk(dev);
1192
1193	if (a == &dev_attr_badblocks.attr && !disk->bb)
1194		return 0;
1195	return a->mode;
1196}
1197
1198static struct attribute_group disk_attr_group = {
1199	.attrs = disk_attrs,
1200	.is_visible = disk_visible,
1201};
1202
1203static const struct attribute_group *disk_attr_groups[] = {
1204	&disk_attr_group,
1205	NULL
1206};
1207
1208/**
1209 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1210 * @disk: disk to replace part_tbl for
1211 * @new_ptbl: new part_tbl to install
1212 *
1213 * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1214 * original ptbl is freed using RCU callback.
1215 *
1216 * LOCKING:
1217 * Matching bd_mutex locked or the caller is the only user of @disk.
1218 */
1219static void disk_replace_part_tbl(struct gendisk *disk,
1220				  struct disk_part_tbl *new_ptbl)
1221{
1222	struct disk_part_tbl *old_ptbl =
1223		rcu_dereference_protected(disk->part_tbl, 1);
1224
1225	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1226
1227	if (old_ptbl) {
1228		rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1229		kfree_rcu(old_ptbl, rcu_head);
1230	}
1231}
1232
1233/**
1234 * disk_expand_part_tbl - expand disk->part_tbl
1235 * @disk: disk to expand part_tbl for
1236 * @partno: expand such that this partno can fit in
1237 *
1238 * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1239 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1240 *
1241 * LOCKING:
1242 * Matching bd_mutex locked or the caller is the only user of @disk.
1243 * Might sleep.
1244 *
1245 * RETURNS:
1246 * 0 on success, -errno on failure.
1247 */
1248int disk_expand_part_tbl(struct gendisk *disk, int partno)
1249{
1250	struct disk_part_tbl *old_ptbl =
1251		rcu_dereference_protected(disk->part_tbl, 1);
1252	struct disk_part_tbl *new_ptbl;
1253	int len = old_ptbl ? old_ptbl->len : 0;
1254	int i, target;
1255	size_t size;
1256
1257	/*
1258	 * check for int overflow, since we can get here from blkpg_ioctl()
1259	 * with a user passed 'partno'.
1260	 */
1261	target = partno + 1;
1262	if (target < 0)
1263		return -EINVAL;
1264
1265	/* disk_max_parts() is zero during initialization, ignore if so */
1266	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1267		return -EINVAL;
1268
1269	if (target <= len)
1270		return 0;
1271
1272	size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1273	new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1274	if (!new_ptbl)
1275		return -ENOMEM;
1276
1277	new_ptbl->len = target;
1278
1279	for (i = 0; i < len; i++)
1280		rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1281
1282	disk_replace_part_tbl(disk, new_ptbl);
1283	return 0;
1284}
1285
1286static void disk_release(struct device *dev)
1287{
1288	struct gendisk *disk = dev_to_disk(dev);
1289
1290	blk_free_devt(dev->devt);
1291	disk_release_events(disk);
1292	kfree(disk->random);
1293	disk_replace_part_tbl(disk, NULL);
1294	hd_free_part(&disk->part0);
1295	if (disk->queue)
1296		blk_put_queue(disk->queue);
1297	kfree(disk);
1298}
1299struct class block_class = {
1300	.name		= "block",
1301};
1302
1303static char *block_devnode(struct device *dev, umode_t *mode,
1304			   kuid_t *uid, kgid_t *gid)
1305{
1306	struct gendisk *disk = dev_to_disk(dev);
1307
1308	if (disk->devnode)
1309		return disk->devnode(disk, mode);
1310	return NULL;
1311}
1312
1313static const struct device_type disk_type = {
1314	.name		= "disk",
1315	.groups		= disk_attr_groups,
1316	.release	= disk_release,
1317	.devnode	= block_devnode,
1318};
1319
1320#ifdef CONFIG_PROC_FS
1321/*
1322 * aggregate disk stat collector.  Uses the same stats that the sysfs
1323 * entries do, above, but makes them available through one seq_file.
1324 *
1325 * The output looks suspiciously like /proc/partitions with a bunch of
1326 * extra fields.
1327 */
1328static int diskstats_show(struct seq_file *seqf, void *v)
1329{
1330	struct gendisk *gp = v;
1331	struct disk_part_iter piter;
1332	struct hd_struct *hd;
1333	char buf[BDEVNAME_SIZE];
1334	unsigned int inflight[2];
1335	int cpu;
1336
1337	/*
1338	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1339		seq_puts(seqf,	"major minor name"
1340				"     rio rmerge rsect ruse wio wmerge "
1341				"wsect wuse running use aveq"
1342				"\n\n");
1343	*/
1344
1345	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1346	while ((hd = disk_part_iter_next(&piter))) {
1347		cpu = part_stat_lock();
1348		part_round_stats(gp->queue, cpu, hd);
1349		part_stat_unlock();
1350		part_in_flight(gp->queue, hd, inflight);
1351		seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1352			   "%u %lu %lu %lu %u %u %u %u\n",
1353			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1354			   disk_name(gp, hd->partno, buf),
1355			   part_stat_read(hd, ios[READ]),
1356			   part_stat_read(hd, merges[READ]),
1357			   part_stat_read(hd, sectors[READ]),
1358			   jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1359			   part_stat_read(hd, ios[WRITE]),
1360			   part_stat_read(hd, merges[WRITE]),
1361			   part_stat_read(hd, sectors[WRITE]),
1362			   jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1363			   inflight[0],
1364			   jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1365			   jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1366			);
1367	}
1368	disk_part_iter_exit(&piter);
1369
1370	return 0;
1371}
1372
1373static const struct seq_operations diskstats_op = {
1374	.start	= disk_seqf_start,
1375	.next	= disk_seqf_next,
1376	.stop	= disk_seqf_stop,
1377	.show	= diskstats_show
1378};
1379
1380static int diskstats_open(struct inode *inode, struct file *file)
1381{
1382	return seq_open(file, &diskstats_op);
1383}
1384
1385static const struct file_operations proc_diskstats_operations = {
1386	.open		= diskstats_open,
1387	.read		= seq_read,
1388	.llseek		= seq_lseek,
1389	.release	= seq_release,
1390};
1391
1392static int __init proc_genhd_init(void)
1393{
1394	proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1395	proc_create("partitions", 0, NULL, &proc_partitions_operations);
1396	return 0;
1397}
1398module_init(proc_genhd_init);
1399#endif /* CONFIG_PROC_FS */
1400
1401dev_t blk_lookup_devt(const char *name, int partno)
1402{
1403	dev_t devt = MKDEV(0, 0);
1404	struct class_dev_iter iter;
1405	struct device *dev;
1406
1407	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1408	while ((dev = class_dev_iter_next(&iter))) {
1409		struct gendisk *disk = dev_to_disk(dev);
1410		struct hd_struct *part;
1411
1412		if (strcmp(dev_name(dev), name))
1413			continue;
1414
1415		if (partno < disk->minors) {
1416			/* We need to return the right devno, even
1417			 * if the partition doesn't exist yet.
1418			 */
1419			devt = MKDEV(MAJOR(dev->devt),
1420				     MINOR(dev->devt) + partno);
1421			break;
1422		}
1423		part = disk_get_part(disk, partno);
1424		if (part) {
1425			devt = part_devt(part);
1426			disk_put_part(part);
1427			break;
1428		}
1429		disk_put_part(part);
1430	}
1431	class_dev_iter_exit(&iter);
1432	return devt;
1433}
1434EXPORT_SYMBOL(blk_lookup_devt);
1435
1436struct gendisk *__alloc_disk_node(int minors, int node_id)
1437{
1438	struct gendisk *disk;
1439	struct disk_part_tbl *ptbl;
 
1440
1441	if (minors > DISK_MAX_PARTS) {
1442		printk(KERN_ERR
1443			"block: can't allocate more than %d partitions\n",
1444			DISK_MAX_PARTS);
1445		minors = DISK_MAX_PARTS;
1446	}
1447
1448	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1449	if (disk) {
1450		if (!init_part_stats(&disk->part0)) {
1451			kfree(disk);
1452			return NULL;
1453		}
1454		init_rwsem(&disk->lookup_sem);
1455		disk->node_id = node_id;
1456		if (disk_expand_part_tbl(disk, 0)) {
1457			free_part_stats(&disk->part0);
1458			kfree(disk);
1459			return NULL;
1460		}
1461		ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1462		rcu_assign_pointer(ptbl->part[0], &disk->part0);
1463
1464		/*
1465		 * set_capacity() and get_capacity() currently don't use
1466		 * seqcounter to read/update the part0->nr_sects. Still init
1467		 * the counter as we can read the sectors in IO submission
1468		 * patch using seqence counters.
1469		 *
1470		 * TODO: Ideally set_capacity() and get_capacity() should be
1471		 * converted to make use of bd_mutex and sequence counters.
1472		 */
1473		seqcount_init(&disk->part0.nr_sects_seq);
1474		if (hd_ref_init(&disk->part0)) {
1475			hd_free_part(&disk->part0);
1476			kfree(disk);
1477			return NULL;
1478		}
1479
1480		disk->minors = minors;
1481		rand_initialize_disk(disk);
1482		disk_to_dev(disk)->class = &block_class;
1483		disk_to_dev(disk)->type = &disk_type;
1484		device_initialize(disk_to_dev(disk));
1485	}
1486	return disk;
1487}
1488EXPORT_SYMBOL(__alloc_disk_node);
1489
1490struct kobject *get_disk_and_module(struct gendisk *disk)
1491{
1492	struct module *owner;
1493	struct kobject *kobj;
1494
1495	if (!disk->fops)
1496		return NULL;
1497	owner = disk->fops->owner;
1498	if (owner && !try_module_get(owner))
1499		return NULL;
1500	kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1501	if (kobj == NULL) {
1502		module_put(owner);
1503		return NULL;
1504	}
1505	return kobj;
1506
1507}
1508EXPORT_SYMBOL(get_disk_and_module);
 
1509
1510void put_disk(struct gendisk *disk)
1511{
1512	if (disk)
1513		kobject_put(&disk_to_dev(disk)->kobj);
1514}
1515EXPORT_SYMBOL(put_disk);
1516
1517/*
1518 * This is a counterpart of get_disk_and_module() and thus also of
1519 * get_gendisk().
1520 */
1521void put_disk_and_module(struct gendisk *disk)
1522{
1523	if (disk) {
1524		struct module *owner = disk->fops->owner;
1525
1526		put_disk(disk);
1527		module_put(owner);
1528	}
1529}
1530EXPORT_SYMBOL(put_disk_and_module);
1531
1532static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1533{
1534	char event[] = "DISK_RO=1";
1535	char *envp[] = { event, NULL };
1536
1537	if (!ro)
1538		event[8] = '0';
1539	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1540}
1541
1542void set_device_ro(struct block_device *bdev, int flag)
1543{
1544	bdev->bd_part->policy = flag;
1545}
1546
1547EXPORT_SYMBOL(set_device_ro);
1548
1549void set_disk_ro(struct gendisk *disk, int flag)
1550{
1551	struct disk_part_iter piter;
1552	struct hd_struct *part;
1553
1554	if (disk->part0.policy != flag) {
1555		set_disk_ro_uevent(disk, flag);
1556		disk->part0.policy = flag;
1557	}
1558
1559	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1560	while ((part = disk_part_iter_next(&piter)))
1561		part->policy = flag;
1562	disk_part_iter_exit(&piter);
1563}
1564
1565EXPORT_SYMBOL(set_disk_ro);
1566
1567int bdev_read_only(struct block_device *bdev)
1568{
1569	if (!bdev)
1570		return 0;
1571	return bdev->bd_part->policy;
1572}
1573
1574EXPORT_SYMBOL(bdev_read_only);
1575
1576int invalidate_partition(struct gendisk *disk, int partno)
1577{
1578	int res = 0;
1579	struct block_device *bdev = bdget_disk(disk, partno);
1580	if (bdev) {
1581		fsync_bdev(bdev);
1582		res = __invalidate_device(bdev, true);
1583		bdput(bdev);
1584	}
1585	return res;
1586}
1587
1588EXPORT_SYMBOL(invalidate_partition);
1589
1590/*
1591 * Disk events - monitor disk events like media change and eject request.
1592 */
1593struct disk_events {
1594	struct list_head	node;		/* all disk_event's */
1595	struct gendisk		*disk;		/* the associated disk */
1596	spinlock_t		lock;
1597
1598	struct mutex		block_mutex;	/* protects blocking */
1599	int			block;		/* event blocking depth */
1600	unsigned int		pending;	/* events already sent out */
1601	unsigned int		clearing;	/* events being cleared */
1602
1603	long			poll_msecs;	/* interval, -1 for default */
1604	struct delayed_work	dwork;
1605};
1606
1607static const char *disk_events_strs[] = {
1608	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
1609	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
1610};
1611
1612static char *disk_uevents[] = {
1613	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
1614	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
1615};
1616
1617/* list of all disk_events */
1618static DEFINE_MUTEX(disk_events_mutex);
1619static LIST_HEAD(disk_events);
1620
1621/* disable in-kernel polling by default */
1622static unsigned long disk_events_dfl_poll_msecs;
1623
1624static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1625{
1626	struct disk_events *ev = disk->ev;
1627	long intv_msecs = 0;
1628
1629	/*
1630	 * If device-specific poll interval is set, always use it.  If
1631	 * the default is being used, poll iff there are events which
1632	 * can't be monitored asynchronously.
1633	 */
1634	if (ev->poll_msecs >= 0)
1635		intv_msecs = ev->poll_msecs;
1636	else if (disk->events & ~disk->async_events)
1637		intv_msecs = disk_events_dfl_poll_msecs;
1638
1639	return msecs_to_jiffies(intv_msecs);
1640}
1641
1642/**
1643 * disk_block_events - block and flush disk event checking
1644 * @disk: disk to block events for
1645 *
1646 * On return from this function, it is guaranteed that event checking
1647 * isn't in progress and won't happen until unblocked by
1648 * disk_unblock_events().  Events blocking is counted and the actual
1649 * unblocking happens after the matching number of unblocks are done.
1650 *
1651 * Note that this intentionally does not block event checking from
1652 * disk_clear_events().
1653 *
1654 * CONTEXT:
1655 * Might sleep.
1656 */
1657void disk_block_events(struct gendisk *disk)
1658{
1659	struct disk_events *ev = disk->ev;
1660	unsigned long flags;
1661	bool cancel;
1662
1663	if (!ev)
1664		return;
1665
1666	/*
1667	 * Outer mutex ensures that the first blocker completes canceling
1668	 * the event work before further blockers are allowed to finish.
1669	 */
1670	mutex_lock(&ev->block_mutex);
1671
1672	spin_lock_irqsave(&ev->lock, flags);
1673	cancel = !ev->block++;
1674	spin_unlock_irqrestore(&ev->lock, flags);
1675
1676	if (cancel)
1677		cancel_delayed_work_sync(&disk->ev->dwork);
1678
1679	mutex_unlock(&ev->block_mutex);
1680}
1681
1682static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1683{
1684	struct disk_events *ev = disk->ev;
1685	unsigned long intv;
1686	unsigned long flags;
1687
1688	spin_lock_irqsave(&ev->lock, flags);
1689
1690	if (WARN_ON_ONCE(ev->block <= 0))
1691		goto out_unlock;
1692
1693	if (--ev->block)
1694		goto out_unlock;
1695
1696	intv = disk_events_poll_jiffies(disk);
1697	if (check_now)
1698		queue_delayed_work(system_freezable_power_efficient_wq,
1699				&ev->dwork, 0);
1700	else if (intv)
1701		queue_delayed_work(system_freezable_power_efficient_wq,
1702				&ev->dwork, intv);
1703out_unlock:
1704	spin_unlock_irqrestore(&ev->lock, flags);
1705}
1706
1707/**
1708 * disk_unblock_events - unblock disk event checking
1709 * @disk: disk to unblock events for
1710 *
1711 * Undo disk_block_events().  When the block count reaches zero, it
1712 * starts events polling if configured.
1713 *
1714 * CONTEXT:
1715 * Don't care.  Safe to call from irq context.
1716 */
1717void disk_unblock_events(struct gendisk *disk)
1718{
1719	if (disk->ev)
1720		__disk_unblock_events(disk, false);
1721}
1722
1723/**
1724 * disk_flush_events - schedule immediate event checking and flushing
1725 * @disk: disk to check and flush events for
1726 * @mask: events to flush
1727 *
1728 * Schedule immediate event checking on @disk if not blocked.  Events in
1729 * @mask are scheduled to be cleared from the driver.  Note that this
1730 * doesn't clear the events from @disk->ev.
1731 *
1732 * CONTEXT:
1733 * If @mask is non-zero must be called with bdev->bd_mutex held.
1734 */
1735void disk_flush_events(struct gendisk *disk, unsigned int mask)
1736{
1737	struct disk_events *ev = disk->ev;
1738
1739	if (!ev)
1740		return;
1741
1742	spin_lock_irq(&ev->lock);
1743	ev->clearing |= mask;
1744	if (!ev->block)
1745		mod_delayed_work(system_freezable_power_efficient_wq,
1746				&ev->dwork, 0);
1747	spin_unlock_irq(&ev->lock);
1748}
1749
1750/**
1751 * disk_clear_events - synchronously check, clear and return pending events
1752 * @disk: disk to fetch and clear events from
1753 * @mask: mask of events to be fetched and cleared
1754 *
1755 * Disk events are synchronously checked and pending events in @mask
1756 * are cleared and returned.  This ignores the block count.
1757 *
1758 * CONTEXT:
1759 * Might sleep.
1760 */
1761unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1762{
1763	const struct block_device_operations *bdops = disk->fops;
1764	struct disk_events *ev = disk->ev;
1765	unsigned int pending;
1766	unsigned int clearing = mask;
1767
1768	if (!ev) {
1769		/* for drivers still using the old ->media_changed method */
1770		if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1771		    bdops->media_changed && bdops->media_changed(disk))
1772			return DISK_EVENT_MEDIA_CHANGE;
1773		return 0;
1774	}
1775
1776	disk_block_events(disk);
1777
1778	/*
1779	 * store the union of mask and ev->clearing on the stack so that the
1780	 * race with disk_flush_events does not cause ambiguity (ev->clearing
1781	 * can still be modified even if events are blocked).
1782	 */
1783	spin_lock_irq(&ev->lock);
1784	clearing |= ev->clearing;
1785	ev->clearing = 0;
1786	spin_unlock_irq(&ev->lock);
1787
1788	disk_check_events(ev, &clearing);
1789	/*
1790	 * if ev->clearing is not 0, the disk_flush_events got called in the
1791	 * middle of this function, so we want to run the workfn without delay.
1792	 */
1793	__disk_unblock_events(disk, ev->clearing ? true : false);
1794
1795	/* then, fetch and clear pending events */
1796	spin_lock_irq(&ev->lock);
1797	pending = ev->pending & mask;
1798	ev->pending &= ~mask;
1799	spin_unlock_irq(&ev->lock);
1800	WARN_ON_ONCE(clearing & mask);
1801
1802	return pending;
1803}
1804
1805/*
1806 * Separate this part out so that a different pointer for clearing_ptr can be
1807 * passed in for disk_clear_events.
1808 */
1809static void disk_events_workfn(struct work_struct *work)
1810{
1811	struct delayed_work *dwork = to_delayed_work(work);
1812	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1813
1814	disk_check_events(ev, &ev->clearing);
1815}
1816
1817static void disk_check_events(struct disk_events *ev,
1818			      unsigned int *clearing_ptr)
1819{
1820	struct gendisk *disk = ev->disk;
1821	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1822	unsigned int clearing = *clearing_ptr;
1823	unsigned int events;
1824	unsigned long intv;
1825	int nr_events = 0, i;
1826
1827	/* check events */
1828	events = disk->fops->check_events(disk, clearing);
1829
1830	/* accumulate pending events and schedule next poll if necessary */
1831	spin_lock_irq(&ev->lock);
1832
1833	events &= ~ev->pending;
1834	ev->pending |= events;
1835	*clearing_ptr &= ~clearing;
1836
1837	intv = disk_events_poll_jiffies(disk);
1838	if (!ev->block && intv)
1839		queue_delayed_work(system_freezable_power_efficient_wq,
1840				&ev->dwork, intv);
1841
1842	spin_unlock_irq(&ev->lock);
1843
1844	/*
1845	 * Tell userland about new events.  Only the events listed in
1846	 * @disk->events are reported.  Unlisted events are processed the
1847	 * same internally but never get reported to userland.
1848	 */
1849	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1850		if (events & disk->events & (1 << i))
1851			envp[nr_events++] = disk_uevents[i];
1852
1853	if (nr_events)
1854		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1855}
1856
1857/*
1858 * A disk events enabled device has the following sysfs nodes under
1859 * its /sys/block/X/ directory.
1860 *
1861 * events		: list of all supported events
1862 * events_async		: list of events which can be detected w/o polling
1863 * events_poll_msecs	: polling interval, 0: disable, -1: system default
1864 */
1865static ssize_t __disk_events_show(unsigned int events, char *buf)
1866{
1867	const char *delim = "";
1868	ssize_t pos = 0;
1869	int i;
1870
1871	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1872		if (events & (1 << i)) {
1873			pos += sprintf(buf + pos, "%s%s",
1874				       delim, disk_events_strs[i]);
1875			delim = " ";
1876		}
1877	if (pos)
1878		pos += sprintf(buf + pos, "\n");
1879	return pos;
1880}
1881
1882static ssize_t disk_events_show(struct device *dev,
1883				struct device_attribute *attr, char *buf)
1884{
1885	struct gendisk *disk = dev_to_disk(dev);
1886
1887	return __disk_events_show(disk->events, buf);
1888}
1889
1890static ssize_t disk_events_async_show(struct device *dev,
1891				      struct device_attribute *attr, char *buf)
1892{
1893	struct gendisk *disk = dev_to_disk(dev);
1894
1895	return __disk_events_show(disk->async_events, buf);
1896}
1897
1898static ssize_t disk_events_poll_msecs_show(struct device *dev,
1899					   struct device_attribute *attr,
1900					   char *buf)
1901{
1902	struct gendisk *disk = dev_to_disk(dev);
1903
1904	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1905}
1906
1907static ssize_t disk_events_poll_msecs_store(struct device *dev,
1908					    struct device_attribute *attr,
1909					    const char *buf, size_t count)
1910{
1911	struct gendisk *disk = dev_to_disk(dev);
1912	long intv;
1913
1914	if (!count || !sscanf(buf, "%ld", &intv))
1915		return -EINVAL;
1916
1917	if (intv < 0 && intv != -1)
1918		return -EINVAL;
1919
1920	disk_block_events(disk);
1921	disk->ev->poll_msecs = intv;
1922	__disk_unblock_events(disk, true);
1923
1924	return count;
1925}
1926
1927static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1928static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1929static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1930			 disk_events_poll_msecs_show,
1931			 disk_events_poll_msecs_store);
1932
1933static const struct attribute *disk_events_attrs[] = {
1934	&dev_attr_events.attr,
1935	&dev_attr_events_async.attr,
1936	&dev_attr_events_poll_msecs.attr,
1937	NULL,
1938};
1939
1940/*
1941 * The default polling interval can be specified by the kernel
1942 * parameter block.events_dfl_poll_msecs which defaults to 0
1943 * (disable).  This can also be modified runtime by writing to
1944 * /sys/module/block/events_dfl_poll_msecs.
1945 */
1946static int disk_events_set_dfl_poll_msecs(const char *val,
1947					  const struct kernel_param *kp)
1948{
1949	struct disk_events *ev;
1950	int ret;
1951
1952	ret = param_set_ulong(val, kp);
1953	if (ret < 0)
1954		return ret;
1955
1956	mutex_lock(&disk_events_mutex);
1957
1958	list_for_each_entry(ev, &disk_events, node)
1959		disk_flush_events(ev->disk, 0);
1960
1961	mutex_unlock(&disk_events_mutex);
1962
1963	return 0;
1964}
1965
1966static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1967	.set	= disk_events_set_dfl_poll_msecs,
1968	.get	= param_get_ulong,
1969};
1970
1971#undef MODULE_PARAM_PREFIX
1972#define MODULE_PARAM_PREFIX	"block."
1973
1974module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1975		&disk_events_dfl_poll_msecs, 0644);
1976
1977/*
1978 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1979 */
1980static void disk_alloc_events(struct gendisk *disk)
1981{
1982	struct disk_events *ev;
1983
1984	if (!disk->fops->check_events)
1985		return;
1986
1987	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1988	if (!ev) {
1989		pr_warn("%s: failed to initialize events\n", disk->disk_name);
1990		return;
1991	}
1992
1993	INIT_LIST_HEAD(&ev->node);
1994	ev->disk = disk;
1995	spin_lock_init(&ev->lock);
1996	mutex_init(&ev->block_mutex);
1997	ev->block = 1;
1998	ev->poll_msecs = -1;
1999	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2000
2001	disk->ev = ev;
2002}
2003
2004static void disk_add_events(struct gendisk *disk)
2005{
2006	if (!disk->ev)
2007		return;
2008
2009	/* FIXME: error handling */
2010	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2011		pr_warn("%s: failed to create sysfs files for events\n",
2012			disk->disk_name);
2013
2014	mutex_lock(&disk_events_mutex);
2015	list_add_tail(&disk->ev->node, &disk_events);
2016	mutex_unlock(&disk_events_mutex);
2017
2018	/*
2019	 * Block count is initialized to 1 and the following initial
2020	 * unblock kicks it into action.
2021	 */
2022	__disk_unblock_events(disk, true);
2023}
2024
2025static void disk_del_events(struct gendisk *disk)
2026{
2027	if (!disk->ev)
2028		return;
2029
2030	disk_block_events(disk);
2031
2032	mutex_lock(&disk_events_mutex);
2033	list_del_init(&disk->ev->node);
2034	mutex_unlock(&disk_events_mutex);
2035
2036	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2037}
2038
2039static void disk_release_events(struct gendisk *disk)
2040{
2041	/* the block count should be 1 from disk_del_events() */
2042	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2043	kfree(disk->ev);
2044}